Logan, John Allen
Working paper no. 1996-07
The two-sided logit (TSL) model proposed in Logan 1996a is conceptually simple but computationally difficult. Conceptually, the TSL model represents occupational outcomes as resulting from mutually constraining preferences and resources of workers and firms. Applying the TSL model gives estimates of the importance of worker and job characteristics in determining employment outcomes. This paper compares the performance of an EM algorithm, the Newton algorithm, and an accelerated EM algorithm (Jamshidian and Jennrich 1993) on the TSL problem. The practical importance of possible local maxima in the TSL likelihood is also considered.